dc.description.abstract | Applications of real-time video encoding, such as video recording and video conference, commonly exist in modern wireless handsets. An H.264 video encoder adopts multiple encoding tools to achieve high coding efficiency at the expense of high computational complexity. The allowable computational complexity for real-time video encoding, however, is generally limited in a wireless handset. Therefore, a complexity control mechanism that well allocates the computational complexity of video encoding under the complexity constraint while maintaining optimal rate-distortion performance is important. This dissertation proposes a Coding-Gain-Based (CGB) complexity control for wireless handsets where video encoding is executed by a general-purpose processor.
In the Coding-Gain-Based (CGB) complexity control, we describe the procedure to obtain the coding efficiency of each encoding tool and present a utilization strategy for these tools. By using the utilization strategy and allocating more complexity to the encoding tools which have higher coding efficiency, the CGB method is able to maximize the overall coding efficiency of the encoder. The complexity overhead of this method is very small, and is a practical method. Though the coding efficiency of encoding tools may vary with different platforms, this research provides a design procedure which allows designers to develop a high efficient complexity control algorithm on different platforms.
Because Motion Estimation (ME), transform, and entropy coding are the most complexity-consuming tools, complexity allocation among these three tools influences the R-D performance of the encoder significantly. This dissertation proposes a Concise-Model-Based (CMB) complexity allocation to obtain the optimal complexity for these three tools. The CMB allocation is composed of two algorithms. The first algorithm performs Complexity Allocation among Encoding Tools (CAET) to allocate complexity to ME, PRECODING, and entropy coding for all MBs in a frame based on a new Complexity-Rate-Distortion (C-R-D) model. This model precisely reveals how ME, PRECODING, and entropy coding influence the C-R-D performance of the encoder with concise formulas. With respect to the model, the algorithm obtains the near-optimal complexity of ME, PRECODING, and entropy coding by a closed-form solution with small complexity overhead. Based on a new C-D model of Motion Estimation (ME), this work proposes the second algorithm that performs Complexity Allocation among Macro-Blocks (CAMB) to further allocate suitable complexity to each MB. Experiments performed on a software-optimized source code show that these two algorithms yield superior performance to the existing algorithms.
| en_US |